All articles
Startups & BusinessNovember 4, 2026 7 min read

The Two-Week Paid Trial: How We Stopped Losing Great Engineers to FAANG Offers

Interview loops lose to salary. A paid two-week trial lets agencies compete on work, autonomy, and team quality — the things engineers actually stay for. Here's how we run ours.

We lost three senior backend candidates in one quarter to product companies paying 30% more. None of them said the money was the real reason. They said they couldn't tell what working with us would actually feel like. That was the moment we rebuilt our hiring loop around a single idea: let the work do the convincing.

Why the standard interview loop fails agencies

Agencies compete in a strange market. We rarely win on base salary against well-funded product companies, and we almost never win on equity upside. What we can offer is variety, autonomy, shorter feedback loops, and the kind of craft that comes from shipping ten different products a year instead of one.

The problem is that none of that is legible in a four-round interview. A system design whiteboard says nothing about whether our codebase is a joy or a swamp. A culture-fit chat with the founder doesn't prove that the PM won't torch the sprint on a Thursday. So the candidate makes the rational choice and takes the offer with the bigger number.

We needed a format that let candidates experience the thing we were actually selling. The answer, after about eighteen months of iteration, is a paid two-week trial. We've now run it with roughly two dozen senior candidates. This is what works, what doesn't, and the parts that still make us uncomfortable.

What the trial actually looks like

The structure is boring on purpose. Boring is what makes it legally and operationally clean.

  • Duration: ten working days, Monday to Friday across two weeks
  • Commitment: part-time is fine (we ask for a minimum of 20 hours), full-time preferred
  • Pay: a flat day rate pegged to the senior contractor band for the candidate's region, paid on a short-form contractor agreement
  • Scope: a scoped slice of real client work, never a take-home puzzle
  • Decision: a yes/no offer by end of day 10, from both sides

Before the trial starts, the candidate has already done a 45-minute intro call and a 60-minute technical conversation. We're not using the trial as a filter for basic competence. We're using it as the final step for candidates we already believe we want to hire.

Picking the right slice of work

This is the part that breaks most attempts at paid trials. If you hand someone a greenfield prototype, you learn nothing about how they operate inside the mess of a real project. If you drop them into the middle of a six-month enterprise build, they spend ten days reading Confluence.

We look for tickets that have three properties:

  1. Real but bounded: a feature that will actually ship, but could reasonably be done by a strong engineer in 3-5 days
  2. Touches more than one layer: ideally API plus client, or infra plus application code, so we see how they think across boundaries
  3. Has at least one ambiguous requirement: because the single most important thing we're hiring for is how someone handles "this spec is wrong, what now?"

A good example from last year: a candidate was asked to add rate limiting to a client's public API, including the decision of whether to do it at the edge or in the application, plus a basic admin view for bypass rules. Bounded, real, and forced a design conversation on day two.

The parts that aren't obvious

Pairing is the actual signal

We require that the candidate pair with at least two different engineers during the trial — typically one on day two or three, and one near the end. Not shadow-pairing. Real pairing on real decisions.

This is where we learn the things interviews can't surface: do they ask good questions, do they argue well, do they update when they're wrong, do they recognise when someone else is wrong without being a jerk about it. After roughly twenty of these, I'd estimate pairing feedback has flipped our decision in about one in four cases, usually from a soft yes to a no.

Give them a boring onboarding

Resist the urge to roll out the red carpet. We give trial engineers the same onboarding doc, the same laptop setup script, and the same Slack welcome message as a regular new hire. If our onboarding is bad, we want them to see it. If it's good, that's a genuine selling point.

# The same script every new engineer runs on day one
./scripts/bootstrap.sh --role=engineer --trial=true
# The --trial flag only scopes credentials, nothing else differs

The one difference: scoped credentials. Trial engineers get access to a single client's staging environment and nothing else. No production, no other client repos, no financial systems. This matters for both security and for the trust conversation we have with the client whose work they're touching.

Tell the client

Speaking of which — yes, we tell the client. Every time. The framing is simple: "We're bringing on a senior engineer and part of our evaluation is having them contribute to real work under supervision. All code is reviewed by a permanent team member before merge. No change to your rate."

In about three years of doing this, exactly one client has said no. Most are mildly flattered that we take hiring seriously enough to do it this way.

What we measure and what we ignore

We deliberately don't score the trial on a rubric. Rubrics in hiring are mostly theatre — they make rejections feel defensible rather than making decisions better. Instead, the two engineers who paired with the candidate and the tech lead on the project each write a short memo answering four questions:

  • Would you want this person on your next project? Why or why not?
  • What did they do that surprised you, good or bad?
  • Where did they struggle, and how did they handle struggling?
  • If we don't hire them, what's the specific reason?

That last question is the one that catches hidden bias. If the only answer is "vibes" or "not quite senior enough" without a concrete example, we push back hard.

The things we stopped caring about

We used to care whether the candidate finished the scoped work inside ten days. We don't anymore. Roughly half don't finish, and finishing turned out to correlate more with scope luck than with engineering quality. We care about the trajectory, the code review conversations, and the design choices — not the burndown.

We also stopped caring about commit volume. One of our best hires wrote maybe 200 lines of code across the trial but asked the question that killed a bad architectural decision on day four. That question saved the project a month of rework.

The uncomfortable parts

Two things about this process still bother us, and I'd rather name them than pretend they're solved.

First, it filters out people who can't take two weeks off from a current job. We try to accommodate — evenings, weekends, split across a month — but a full-time employed candidate with kids and a mortgage is at a real disadvantage versus someone between jobs or freelancing. We've started offering a compressed one-week version for these cases, which is worse but better than nothing.

Second, a paid trial is not the same as a job offer, and some candidates understandably won't do it. Our acceptance rate for the trial invitation sits somewhere around 60-70%. The ones who decline usually have a competing offer already in hand and no time. We've accepted that we'll lose some good people to this and decided the tradeoff is worth it, because the hires we do make through this process stay longer and ramp faster. In our experience the first-year attrition on trial hires is roughly half that of our old interview-only hires, though the sample is still small enough that I wouldn't bet a company on that number.

Where we'd start if we were doing this from scratch

If you run an agency and you're tired of losing engineers to bigger offers, don't try to redesign your whole hiring loop at once. Pick your next senior backend or full-stack hire. Keep your existing intro and technical calls. Replace the final round with a paid five-day trial on one specific, bounded client ticket. Require pairing on at least two of those days. Write the four-question memo at the end.

You'll learn more about the candidate in those five days than in six rounds of interviews, and they'll learn more about you. Sometimes that means they'll say no. That's the point — a no in week one is cheaper for everyone than a resignation in month four.

If you want to see how we structure the contractor agreement or the client disclosure template, we've written about adjacent hiring practices over on our blog and talk about team composition in our engineering services pages.

#hiring#agency#engineering culture#recruitment

Want a team like ours?

72Technologies builds production software for the kind of teams who actually read this blog.

Start a project